多层感知器在机械故障诊断中的应用

Application of the multi-layer perceptron in the machinery fault diagnosis

提出一种动态调整学习率和附加梯度变化量与动量项相结合的权值优化方法,同时引入绝对误差函数用于对多层感知器中BP算法的改进,并将改进算法用于旋转机械故障诊断实例样本的学习。仿真结果表明,改进的BP算法可显著加速网络训练速度,学习过程具有较好的收敛性,并能正确地诊断出存在的故障,具有一定的实用价值。

?A optimal method with gradient variance,momentum items and adaptive learning rate factors is presented for the modification of weights;Meanwhile the absolute error function is also added for the improvement of back propagation(BP) algorithm.The improved algorithm is applied to the learning process of the fault diagnosis for the rotating machinery.The simulation results show the presented quick training algorithm can speed up the learning process of multilayer perceptron,and improve the learning properties on convergence and robust performance.The method is effective to the diagnosis of the mechanical faults.